Model Square

Creates smooth, high-quality image-to-video results while preserving the source image.

Supports multiple reference images while preserving subjects, scenes and creative intent.

Understands text prompts accurately and generates smooth, detailed, high-quality video.

Edits video through natural language and reference images while accurately reproducing motion.

Creates smooth, high-quality image-to-video results while preserving the source image.

Supports multiple reference images while preserving subjects, scenes and creative intent.

Understands text prompts accurately and generates smooth, detailed, high-quality video.

Kling 3.0 high-resolution image-to-video generation.

Kling 3.0 high-resolution text-to-video generation.

字节 Seedance 1.0 Pro：文生 / 图生 / 参考图 / 首尾帧四模式，5–10 秒，480P/720P/1080P，按秒计费。

字节 Seedance 1.5 Pro：四模式，4–12 秒，480P/720P/1080P，按秒计费。

字节 Seedance 2.0：四模式，4–15 秒，480P/720P/1080P，原生音频可开关，按秒计费。

ByteDance Seedance 2.5: text / first-frame / reference / first+last-frame modes, 4–30 s, 480P–1080P, native audio switchable, billed per second.

Stable image-to-video generation with strong instruction following, camera control and visual consistency.

Stable text-to-video for large complex motion, cinematic visuals, strong instruction following and physical realism.

Wan 2.6 image-to-video supports 10- and 15-second output at 720P or 1080P.

Wan 2.6 text-to-video supports 10- and 15-second output with improved image quality.

Wan 2.7 image-to-video supports first-frame, first-and-last-frame and video-extension modes at 720P or 1080P.

Wan 2.7 text-to-video supports resolution and aspect-ratio combinations with 5–15 second output.

Alibaba Wan 3.0 all-in-one video model: text / first-frame / reference / first+last-frame modes, 2–30 s, 480P–1080P at 30fps, native audio switchable, billed per second.